ECG Signal Analysis for Automatic Cardiac Abnormality Detection and Classification

Abstract

Electrocardiogram (ECG), a noninvasive system that is used as a crucial diagnostic tool for cardiovascular diseases. A prepared ECG signal provides indispensable information about the electrophysiology of the heart diseases and cardiovascular changes that may occur. Also it offers valuable information about the functional characteristics of the heart and cardiovascular system. When monitoring ECG for a long period of time about 24 hours it is tedious, so because of that the optical analysis cannot be trusted upon and the possibility of the analyst missing the dynamic information is high. So, computer based investigation and classification of diseases can be very supportive in diagnosis of cardiovascular diseases (CVD). This research was able to develop a system for ECG signal analysis that will analyze the signal with a good, quality and precise feature extraction and classification of ECG wave form to detect diverse heart disease complications. From the literatures, it was point out that the ECG analysis systems established by using hybrid algorithms are too difficult. But, the hybrid techniques that have been applied in the researches yields improved analysis of heart disease classification. This research is implemented using Discrete Wavelet Transform (DWT) and Principal Component Analysis (PCA) for feature manipulation and Adaptive Neuro Fuzzy Inference System (ANFIS) as a Neuro Fuzzy classifier in classifying Normal, Left Bundle Branch Block(LBBB), paced beat, Right Bundle Branch Block(RBBB) and Supraventricular Contraction(SVC) of ECG signals. The research used physionet database with labelled ECG signals with different cardiac problems. From those data’s using DWT it was able to extract around six features and due to inefficiency of the machine processor it was reduced using PCA into five vital feature vectors. Then taking only the detail D4 level decomposition of each signals for calculating the features and feeding into ANFIS classifier it was made possible to attain an overall accuracy of the system about 99.34% with average of 99.36% and 99.84% sensitivity and specificity respectively.